General discussions on control systems Chapter | 1 41

similar to level 2 only difference here the coordination 1.3 Control system models
among plants instead of units. An overall production
scheduling and control activities so as to get optimum 1.3.0 Discussions on control system models
solutions for profitability, quality control including For controlling a process it is essential to understand the
meeting requirements of local standards, savings in process and its dynamics. Process dynamics refers to the
cost (toward inventory, raw material supplies and trans- time varying behavior of the process when it is subjected to
portation), energy and time. Such network could be pri- various external and internal disturbances. Model based
vate or may use public platform also. It also takes care control is smart system, when compared to typical con-
of availability, reliability and overall diagnostics. Let us trollers, to offer better disturbance rejection and uniform
take an example: there is an order of type 5 cement, the control including set point tracking. However, to imple-
authority has (say) 3 plants, in one there will be annual ment the same higher skill and understanding of the process
maintenance, other is busy with type 3 production, third and operation of the plant is essential. Also, it requires
one is available for timely production but has lesser ca- mathematical modeling and high computing power of the
pacity. Naturally production scheduling of sharing pro- control systems. Therefore, such models are implemented
duction at two plants and make due coordination for in DCS/PLC systems. First let us look into the basic aspects
getting it produced at two plants. All such supervision of process dynamics and terms frequently used in control
and coordination activities will be carried out from system modeling:
this level.
6. Level 4: This is mainly responsible for management in- 1. Parameters: Lumped and distributed parameters are
formation of the entire set ups of the company and represented by differential equation and in partial
keeping all information ready as executive summary different equations respectively.
to the top management group. It also functions as intra- 2. Linearity, stability and order of system: Linear sys-
company communication for sales order, basic manage- tems have linear functions with homogeneity and super-
ment, materials management etc. This also includes position (possible). System stability and order etc. refer
enterprise computer network duly connected through I 1.0.5.
internet (say) or dedicated company network. Also, 3. Types of transfer function concepts: Model based
when people going in for Industry 4.0 and digital trans- control (MBC) or Model based design (MBD) mainly
formation, there will be immense need for IT data ex- stands on the pillar of mathematical computations.
change to strategize operating technology for better Transfer functions and their relations as have been dis-
production efficiencies. Since there will be chances of cussed in I 1.0.5 and detailed out in Fig. I 1.3a play
external access at some points in the network, cyber se- important role in developing model based control
curity is very importance. So, Firewall and cyber secu- systems.
rity at various level is essential and symbolically shown 4. Process dynamics variations: There could be various
at the top level. types of dynamics possible in a process such as: Instan-
taneous process: Y(t) ¼ K X(t) where K is the system
Let us now look into modeling control systems.
static gain or the process could be integral type such

FIGURE I 1.3a Mathematical model of control loop.

42 Plant Intelligent Automation and Digital Transformation

that dY/dt ¼ X example a liquid level container with the system i.e., system input x1 will produce output
more inflow than outflow to give an example of positive y2, which is not proportional x1.
integral slope. Also, processes may be First/second/ 5. Characteristics of nonlinear system: Major issue
higher order typesdrefer I 1.0.5. Some processes are related to nonlinear system are:
self-regulating type or could be process with delay. l Superposition: The superposition principle is not appli-

5. Simplicity versus Accuracy: For a mathematical cable for nonlinear system. Let us consider two exam-
model, two issues of simplicity and the accuracy, which ples, Y ¼ X and Y ¼ X2 in former case if X ¼ 1;
may be opposite in nature need to be considered. For Y ¼ 1, X ¼ 2; Y ¼ 2 X ¼ (1 þ 2) ¼ Y ¼ (1 ¼ 2) ¼
simplicity, at times it may be necessary to ignore 3, but in Second case X ¼ 1 Y ¼ 1; X ¼ 2 Y ¼ 4,
some minor issues/parameters and the result from the but X ¼ (1 þ 2) ¼ 3 BUT Y is NOT EQUAL 5 but
model may not be accurate but a good one. It is always 9dmeaning it does not follow superposition.
better to start with simplest model to get the feeling of l Equilibrium points: The nonlinear systems have

the system then complete mathematical model may be several equilibrium and stable points with defined
built for more accurate analysis. Before mathematical cycles.
model, let us look into various types of systems. l Controls: Nonlinear systems can be decomposed

into multiple linear systems. Major control types

1.3.1 Types of systems used for nonlinear systems are: PID Control/Model
based control, Local linearization and Gain Sched-
There could be different forms for mathematical models uling, Feedback Linearizationdstate variable con-
based on process dynamics. Most models applied to a trol, Mode Control, Optimal control, Model
working plant use actual historical data combined with predictive control, Adaptive control, Neural network
basic stoichiometric relationships e.g., state space approach control and Fuzzy control etc.
is better suited for optimal control applications. So, ex-
change of historical and other IT data for operational model
developments are very helpful for Industry 4.0 approach. 1.3.2 Model based control (MBC) methodology
Depending on the individual system and the given cir- and structure
cumstances, one particular type of mathematical model MBC methodology and structure are:
suits better than other models.
1. MBC Methods (mathematical): It is beyond doubt that
1. Linear Systems: A system is referred to as linear when making experiments is the best way to collect data
the principle of superposition could be applied. Accord- about a process. Time, effort and cost associated with
ing to principle of superposition, when two or more experimental approach, may not be an easy option,
different input functions are applied simultaneously to especially for cost part may prohibit carrying out real
a system, the response produced is the sum of the two experiment viz. utility boiler. Mathematical model is
individual responses due to these input functions. one well accepted method. Mathematical model are
Hence, in a linear system, the response for several in- developed to describe physical and chemical phenom-
puts can be calculated by treating one input at a time ena involved in a process. However, it should be borne
and adding the results. Thus, it is possible to get compli- in mind that though it is a good alternative yet, it cannot
cated solution to linear differential equations from indi- be perfect alternative to real process, with some scope
vidual simple solutions. So, a differential equation is of error. Efforts should be there to minimize this error
linear when it has constant coefficient or the coefficient as far as possible. Mathematical model can be two types
is a function of independent variable i.e., when a 0, a1 viz. Mechanistic model and Black-box model.
etc. in Eq. I 1.0e, satisfy the aforesaid criteria. l Mechanistic Model: In Mechanistic model the dy-
2. Linear Time invariant Systems: In linear time namic behavior of the process is described by the
invariant dynamic systems, the coefficients of the afore- set of differential equations. Mechanistic model is
said differential equations are constant (a0 or a1 are con- used for a completely known process, where the char-
stants). These systems have linear time-invariant acteristics of the process are well defined. The mech-
lumped-parameter components. anistic model is again divided into a lumped
3. Linear Time variant Systems: Linear systems with parameter or a distributed parameter described by or-
aforesaid differential equations where coefficients are dinary and partial differential equations respectively.
functions of time are called linear time-variant systems. These lumped and distributed parameter models could
4. Nonlinear system: In nonlinear system the output from be linear or nonlinear. Nonlinear model are expressed
the system does not vary proportionately with input to as piecewise linear systems with transition points.
 General discussions on control systems Chapter | 1 43

l Black box: Black box is second category of mathe- possible for the model to predict the effects of varying
matical model. It is expected to develop a transfer process operating conditions in terms of control re-
function-based model by performing a plant test. In quests. In the model-based controller designs, simple
this method, first bring the process to a required models like First/Second order process with delay
steady-state operating point then make a step change time (FOPDT/SOPDT) are often used to describe the
in the manipulated input (i.e., controller output) and dynamics of a given process.
the process response is observed (Refer Fig. I 1.3b). 2. MBC Structure: Based on suitable, a process model
Finally, a model is recognized to bring the greatest acting as controller is developed in suitable way, to
match between the observed plant output and the calculate the manipulated variable, which should force
model output. Black box type has a drawback, that the controlled variable (x) to behave in a desired way,
the parameters of these transfer functions do not so that the calculated value of m can be implemented.
have any physical implication in terms of similarity Thus, such controller without PID elements, there is
to process parameters. But the black box modeling only one parameter per controlled variable. In MBC
method is effective, if the purpose is to denote truly as discussed, model determines the process input that
some developments in the process behavior. causes a desired process output response from the pro-
In transfer function based mathematical model normally cess. So, the model is referred to as inverse controller
use open loop (Refer: Fig. I 1.3a) Laplace transform for designated as CI [4]. In practical applications on ac-
step input responses for analysis. Time series models count of disturbances, an open loop discussed, will
represent the open-loop response of the process with a not provide desired result so, feedback with adjustment
vector of impulses that are empirically determined [4]. mechanism should be adapted. In this configuration, CI
Often nonlineardModel predictive control (MPC) has been provided with feedback from adjustment
with valid for the operating range, is used. In all such mechanism CA. CA acts on the difference
cases there will be multiple linear models having transi- (d) produced by actual process and model as depicted
tions (between these linear models based on operating in Fig. I 1.3b (from Chapter 2_13 of [4]).
region). To deal with transition issues gain scheduling,
neural network based models and other methods like 1.3.3 Various model based control types
Fuzzy logic expert rules could be used. Nonlinear
phenomenological models are design-type simulators Depending on the model type various model-based
[4]. Irrespective of type of mathematical model used, control are: Process Model Base control (PMBC), Model
there should be facilities scaling in terms of time and Predictive control (MPC) and Internal Model control
space. Also, with given information, it should be (IMC).

FIGURE I 1.3b Model based control concept.

44 Plant Intelligent Automation and Digital Transformation

1. Process Model Base Control (PMBC): Process control action the controller will correct the 25% error
models in PMBC are normally developed from a mech- by 75%. In this manner the deviation could be as close
anistic type on the basis of balancing act on material to 0.25n after n numbers of control action [16]. As
and energy with various objectives and other desirable shown in the figure there are two sets of modules:
issues, such as capital cost, flexibility, resource use, en- the Physical model base (PMB) module and PMB
ergy integration. Here often sustainability is compro- controller. PMB module (correction module) has a
mised for accommodating process control. For model of the unit to be controlled with same inputs
nonlinear processes, gain scheduling is typical conven- as that for the process. The purpose of this module is
tional solution, whereas, in PMBC engineer’s process to correct, the prediction made by the controller mod-
knowledge is expressed as a dynamic nonlinear model, ule. PMB Controller actually predicts the value of the
which is also comparatively simpler. Simplest dynamic manipulated variables, to get the desired performance,
model for a typical Single input single output (SISO) for given measurable disturbance as well as for
can be represented by: dx dt ¼ f(y, u, d, p) where p is changes in set point taken as input to PMB controller
adjusted process parameter, while other parameters are module. The output from PMB Controller goes as
as shown in Fig. I 1.3c for PMBC. Model parameters input to the process only after being corrected by con-
are chosen meticulously so that it can represent the pro- trol variable generated by the process and PMB mod-
cess in the question. Therefore, the model parameters ule (used as correction adjustor). So, same
should be realistic, the chosen mechanism has the flex- disturbance is supplied to process and PMB module
ibility for changes in such a way it has direct influence to generate control variables (x) designated as xp and
on the process output. PMBC has several advantages xm respectively, and these two are used to correct
such as single tuning parameter, nonlinear compensa- manipulated variable under changed circumstances
tion over the entire operating range, preserves process (Changes in set point or disturbance rejection) and
knowledge, and provides continuous monitoring of are used as feedback.
the process [16]. This can be used in some predictive 2. Model Predictive Control: Model predictive control is
control and for Multi input Multi (/Single) output very popular because it can be applied to Multiple-input/
(MIMO/MISO). output (MIMO) process, while satisfying inequality con-
Model system may not be 100% perfect but could be straints on the input and output variables. For applicability
quite good to attend the goal very close to desired con- in MIMO, here the changes in each of input variables is
dition. For control, the model perfection is not manda- duly coordinated after establishing the input-output rela-
tory; even if there is error as high as 25% in next tionships represented by model. Refer Fig. I 1.3d.

FIGURE I 1.3c Process model base control

scheme.
 General discussions on control systems Chapter | 1 45

FIGURE I 1.3d Model predictive control scheme.

In MPC applications, CV, MV and measured distur- l Conceptual idea: MPC is a control type where cur-
bance variables (DV) are well coordinated. rent control action is calculated by solving, an (finite
l Advantages: Major advantages are: MOC captures horizon) open-loop optimal control problem, using
and coordinates among the dynamic and static inter- the current state of the plant as the initial state; the
actions of MV, CV and DV including associated optimization yields an optimal control sequence
constraints in systematic way. It can predict any po- and the first control in this sequence is applied to
tential problem in advance and has the capability of the plant [17]. For this, a reasonably accurate dy-
system optimization to produce optimum output. namic model of the process is made available, model
46 Plant Intelligent Automation and Digital Transformation

and current measurements can be used to predict in Fig. I 1.3d3. As indicated above, upper and lower
future values of the outputs. Through feedback the limits, can be included in set point and control cal-
difference between predicted and actual input vari- culations (for inequality constraints). So, we see
ables, the changes may be taken care of. Thus, that MPC calculations are based on current mea-
through various stages of sampling and corrections surements with future predictions for output. The
desired set point/target is achieved. As it deals with objective of the MPC control calculations is to
open loop process response just like vectors it is determine a sequence of control moves (i.e., manip-
often called Time series control. The basic concept ulated input changes) so that the predicted response
is detailed our in Fig. I 1.3d 1 and 3. Future error moves to the set point in an optimal manner [18].
along with various inequality constraints are applied 3. Internal Model control (IMC): IMC philosophy,
to optimizer to modify optimizer output. Optimizer which uses open loop response of Laplace transfor-
output is applied to dynamic model as well as to mation, has been developed on the model principle:
the actual process. Since model also accepts the “Accurate control can be achieved only if the control
actual process, it is in a better position to predict systems encapsulate (either implicitly or explicitly)
new output under theses constraints and process con- some representation of the process to be controlled”
ditions. The new predicted output from the model in [19]. As IMC, shown in Fig. I 1.3e, is based on
conjunction with target set generates the new future pole zero cancellation, methods, IMC controllers
error for the optimizer. The targets, are calculated offer good set-point tracking and disturbance
from an economic optimization based on a steady- response for the process especially with a small-
state model of the process [18]. Optimization objec- time delay. However, in some cases disturbance re-
tives could be maximization of profit/production or jections are of prime importance than set point
minimization of cost as shown. Also, optimum set tracking; especially for unstable systems. In Smith
points change frequently due to variations of process predictor (detailed out later in this book) IMC is
conditions. There could be changes in constraints used to compensate for disturbances due to delay.
from process conditions, equipment (say-reaching Therefore, due consideration must be given while
limit), and instrumentation (say failure), in addition selecting the controller type. Disturbances in the sys-
to economic data discussed above. tem can be due to, noise and model uncertainty. IMC
l System details: In MPC, calculation of set point is offers long settling time due to load disturbances. So,
carried out each time with the control calculations while applying this, due consideration for the same
are performed. This is better detailed out shown in should be borne in mind.
Fig. I 1.3d2, to show how difference between model In this scheme there is a (low pass) filter at the begin-
and actual process output is used as feedback to ning. Performance and parameterization of IMC
modify the changes in set point and control calcula- controller is dependent on the IMC filter time constant.
tions through prediction block. Prediction Higher the filter time constant lower will be the over-
and control action horizon have been detailed out shoot but at the same time lower may be the disturbance

FIGURE I 1.3e Internal model control scheme.

 General discussions on control systems Chapter | 1 47

rejection. Laplace transforms are used here to deal the rejections can be achieved. However even with G e p(s)
issues with transfer functions. For basic analysis, let s Gp(s) disturbance rejection can be achieved as long
the transfer functions of controller and the process be as Gc(s) ¼ G e p (s)L1 condition is satisfied. As stated
Gc(s) and Gp(s) respectively. When controller is set con- above, for better performance of IMC a low pass filter
trol on the plant Gp(s) with set input r(s), then, for a is necessary. So, if Gf(s) be the transfer function of fil-
linear system the output Y(s) will be given by: ter, then internal model controller transfer function can
be conceived as
YðsÞ ¼ rðsÞ$Gc ðsÞ$Gp ðsÞ (I 1.3a)
Gimc ðsÞ ¼ Gc ðSÞ.Gf ðsÞ (I 1.3f)
When a model of the plant with transfer function is
modeled as G e p ðsÞ such that G
e p(s) ¼ Gp(s); then for set Therefore, by combining (Eq. I 1.3e) and (Eq. I 1.3f)
point tracking; one would get
1
h i
e p ðsÞ
Gc ðsÞ ¼ G (I 1.3b) e
GimcðsÞ:rðsÞ:GpðsÞ þ 1 GimcðsÞ:GpðsÞ :dðsÞ
yðsÞ ¼ h i
This is open loop but, for real life applications, process e
1 þ GpðsÞ GpðsÞ :GimcðsÞ
models have mismatch with the parent process; hence
(I 1.3g)
feedback control schemes are designed to counteract
the effects of mismatching. Internal Model Control For linear process model IMCs provide good robustness
(IMC) scheme is a simple scheme, which can be built against disturbances and model mismatch.
from the ideas implemented in the open loop model
strategy discussed and shown in Fig. I 1.3e where, the
process model G e p ðsÞ plays an important role. For the 1.3.4 Control computing and system
e
ideal case Gp ðsÞ ¼ Gp(s); the feedback is only affected development
by disturbance d(s) thus, system is effectively open After gathering some knowledge on modeling, it will be
loop and hence no stability problems can arise. The good idea to know briefly about control computing and
manipulated input u(s) is introduced to both the process control system development. These control computations are
and its model. The process output y(s) is compared with carried out different platforms such as in Advanced appli-
the output of the model resulting in cation process control modules in DCS/PLC, Work station,
h i computing server and specialized controllers. In the modern
b
dðsÞ ¼ Gp ðsÞ GðsÞ e .uðsÞ þ dðsÞ (I 1.3c) era, Application specific integrate circuit (ASIC), micro-
processor with software, Digital signal processing (DSP),
Now, if d(s) ¼ 0 then it is the difference in behavior be- microcontroller or Field programmable gate array (FPGA)
tween actual process and model so with the assumption and system on chip (SoC) are there to implement these.
e p (s) ¼ Gp(s), b
G d (s) stands to represent unknown distur- Control development cycle normally includes: Simplified
bance or noise. Thus, b d (s) can be used to improve the design modeling (Physical/empirical etc.) analysis and sys-
control and may be treated as the missing information in tem design. Next stage could be simulation (especially for
the model. This can be achieved by sending an error software development e.g., MATLAB/Simulink) for design
signal to the controller. The error signal incorporates validation and verification with basic specifications. Final
the model mismatch and disturbances and helps to stage could be commissioning and certification. Based on
achieve the set-point. Hence Manipulated variable u(s) requirement specification for control system, various stages
will be given by: of control system development can be divided into four
½RðsÞ dðsÞ GcðsÞ stages viz. Analysis, Simulation, validation/verification and
uðsÞ ¼ h i (I 1.3d) deployment. In each stage, there are subdivisions also. These
e
1 þ GpðsÞ þ GpðsÞ GcðsÞ stages along with their subdivisions have been depicted in
Fig. I 1.3f. The figure is self-explanatory.
As from Fig. I 1.3e one gets that y(s) ¼ u(s). Gp(s) þ
d(s); So, 1.3.5 Linearization of nonliner systems
½RðsÞ dðsÞ GcðsÞ
yðsÞ ¼ h i :Gp ðsÞ þ dðsÞ (I 1.3e) Recalling previous discussions, we know that in a linear
e
1 þ GpðsÞ þ GpðsÞ GcðsÞ system, superposition is applicable. On the contrary, in case
of nonlinear system the principle of superposition, does not
If G e p ðsÞ 1 conditions are
e p ðsÞ ¼ Gp(s) and Gc(s) ¼ G apply. Although in many applications actually systems are
satisfied, then set point tracking and disturbances actually nonlinear but through careful study by piecewise
48 Plant Intelligent Automation and Digital Transformation

FIGURE I 1.3f Control analysis and development process.

linearization systems are treated as linear system. Here curve of the system can be considered as linear. This
short discussions on some mathematical tools will be has been depicted in Fig. I 1.3g. In this figure let the
recapitulated. nonlinear function be y ¼ x2 þ 1. Linearizing this
1. Concept of linearization: From the above it is clear nonlinear function about the operating point x ¼ 1,
that even so-called “linear systems” are really linear y ¼ 2 as shown, results in a linear function y ¼ 2x.
only in limited operating ranges i.e., around operating This means that near the operating point, y ¼ 2x is a
point, now, within limited operating range nonlinear good approximation for the nonlinearized curve, but
 General discussions on control systems Chapter | 1 49

FIGURE I 1.3g Linearization of nonlinear function.

this is possible only over a limited operating range around an equilibrium point, and the signals may
around the operating point and beyond this such linear- be considered small signals around the equilibrium
ization would give poor result. This goes to establish it barring some special exceptions. So, if the signals
is possible to linearize through piecewise linear curves. involved are small signals, then within a limited oper-
For this let us consider x(t) as status of the system, u(t) (for ating range, the nonlinear system could be approxi-
controller it could be error, which at equilibrium will be mated by a linear system. This is possible by Taylor
zero!) as input and y(t) be the output such that differential series about the operating point and the higher order
equation below can be used for linearization: terms could be neglected (slight deviation from oper-
ating point so that small enough to be neglected) on
xðtÞ
_ ¼ f½xðtÞ; uðtÞ; t (I 1.3h)
account of small signal values.
yðtÞ ¼ g½xðtÞ; uðtÞ; t (I 1.3i) Let the system where input is x(t) and output is y(t) i.e.,
simply y, x such that
To represent the linearized model, define new variables
y ¼ fðxÞ (I 1.3k)
centered about the operating point (represented by suf-
fix 0) such that at operating point: t ¼ t0, x(t0) ¼ x0, If normal operating condition exists at (x0, y0).
u(t0) ¼ u0, and y(t0) ¼ g(x0, u0, t0) ¼ y0. For this we Then by Taylor series, Eq. I 1.3k can be represented by
need to define new variables around operating point
such that: dx(t) ¼ x(t)-x0du(t) ¼ y(t)-y0. The linearized df 1
y ¼ fðx0Þ þ :ðx x0 Þ þ :ðx x0 Þ2 (I 1.3l)
model in terms of dx, du, and dy is valid when the dx 2!
values of these variables are small: where, the derivatives are evaluated at x ¼ x0. If the
_
dxðtÞ ¼ þ A.dxðtÞ þ B.duðtÞ:dyðtÞ ¼ CdxðtÞ þ DduðtÞ variation is very small, we may neglect the higher-
order terms involving (x x0). Then Eq. I 1.3l can be
(I 1.3j) rewritten as
2. Linearization and Taylor series: From above it is ðy y0 Þ ¼ Kðx x0 Þ (I 1.3m)
clear that a normal operation of the system may be df
Where f(x0) ¼ y0; and K ¼ dx at x ¼ x0
50 Plant Intelligent Automation and Digital Transformation

Equation gives a linear mathematical relation for the 1.4.1 State variable and process
nonlinear system represented by Eq. I 1.3k near oper-
The state variables are primarily the fundamental quantities
ating point. In the similar manner this can be extended
of a process viz., mass, energy and momentum, whereas the
for a system with more than one input (say) two inputs
state equations are generated out of conservation principle
x1 and x2 i.e., for
involving these fundamental quantities. Let A be the
y ¼ fðx1; x2Þ ðnonlinear functions of two variablesÞ fundamental quantity accumulation. If I/O ¼ Incoming/
(I 1.3n) Outgoing flow rate with respect to system respectively;
G/C ¼ rate of quantity Generated/Consumed in the system.
Then such linear relationship could be: Then state equation can be:
ðy y0 Þ ¼ K1 ðx1 x10 Þ þ K2 ðx2 x20 Þ (I 1.3o) Rate of accumulation of A ¼ I þ G ðO þ CÞ (I 1.4a)
Where x10, x20 are value of x1 and x2 at operating point Let there are n no. of streams going inside a process and
respectively. m no. of streams coming out of a process. In this system let
All the above linearization equations are valid in near us apply conservation of mass and energy to find out mass
the vicinity of the operating condition and the operating and energy balance. Let r, V represent density and volume
conditions do not vary widely. of material in the system. If Fi,j and ri,j represent incoming
3. Application area: For majority of control functions and (i) and outgoing (j) volume flow and density respectively
analysis it is necessary to have linear equations. Linear- and let h and hi,j represent the energy (enthalpy) in the
ization finds major usage in the following cases: process and in incoming and outgoing streams. Also let Q
l Loop analysis and loop responses at different oper- be the energy addition/deletion to the system and W work
ating points done by/on the system. From these quantities one can find
l Development of linear control system out.
l Plot the Bode response (including in simulation).
Mass Balance: Mass balance with time for the system
can be carried out by:
l Loop stability margins and open-loop response.
l Development of loop frequency responses drV Xn n
X
¼ :Fi. ri :Fj. rj (I 1.4b)
l Development of control system model with less dt i¼1 i¼1
model error
With knowledge gathered, now, let us look into another Energy balance: Energy (E) balance with time for the
control means based on state space system, which is system can be found by:
very important for control system study. dE Xn X
n
¼ :Fi. ri. hi :Fj. rj. hi Q W (I 1.4c)
dt i¼1 i¼1
1.4 State space in control systems
In the similar way component balance due to chemical
[Also refer I 13.2.2 for analysis]. reactions in a chemical process can be carried out when
concentrations of reactants are known. All these are ex-
1.4.0 Concept of state space and control amples of state equations, which are formed by internal
system variable to indicate internal conditions for mass/energy/re-
action conditions etc. Therefore, these are state-space
In contrast to the classical control theory, which basically description of the process to provide the dynamics of the
provides input-output description of the system normally system as a set of coupled first-order differential equations
expressed in terms of transfer function, modern control in a set of internal variables known as state variables.
theory of state space approach provides the state space Thus, the state of the system can be conceived as a vector
description of the system also. In state space approach the within the specified space.
system dynamics are expressed as a set of coupled first-
order differential equations in a set of internal variables
1.4.2 System state
known as state variables, along with a set of algebraic
equations to combine the state variables with output vari- Definition of state of system: There could be several sets
ables. The state variables are a set of fundamental depen- of variables, which can describe dynamics system. The
dent quantities whose values describe the natural state of a concept and description of state of a dynamic system
given process. The state equations are set of differential refers to a minimum set of variables, that can completely
equations, which describe the progression of the states with describe the system along with its response to any given
time. set of inputs is known as state variables. Mathematically,
 General discussions on control systems Chapter | 1 51

complete description of the system in terms of a mini- Y ¼ CX þ DU (I 1.4e)

mum set of variables (say) xi(t), i ¼ 1, ., n, together
with knowledge of those variables at initial time t ¼ t0 Here state variables X (x1, x2, xn) represent state vector, in-
and the system inputs for time t t0, are sufficient to puts U (ui, u2, ur) represent input vector and Y (y1,y2, ym)
predict the future system state and outputs for any time output. Here Matrices A and B are termed as state and input
t > t0. Fig. I 1.4a State Space of System. This ensures matrix respectively. Matrices C and D are termed as output
that the dynamic behavior of a state-determined system is matrix and transition matrix respectively.
completely described by the response of the set of n
(minimum) variables xi(t), where the number n indicates 1.4.3 State equation
the order of the system. So, for a second order system, a
set of at least two variables is necessary to describe the This standard form of the state equations is used throughout
dynamics of the system with known initial conditions of to express system dynamics. This is a form the mathe-
the variables. Fig. I 1.4a1 shows a generalized system matical description of the state equations expressed as a set
with “n” number of states where “r” numbers of input of n coupled first-order ordinary differential equations
vectors are placed to system to produce “m” numbers of where the time derivative of each state variable is expressed
output. The state variables describe internal condition to in terms of the state variables x1(t), ., xn(t) and the system
characterize the system completely at any time t. from inputs u1(t), ., ur(t).
these equations output variables yi(t) could be computed. So, equations involving derivatives of x state informa-
In connection with mass and energy balance, it has been tion in terms of x and input u is referred to as state equation.
seen that derivative of state variables are first computed. This equation basically it means:
State equations: The two basic equations of state vari- x_ 1 ¼ f 1 ðx; u and tÞ
ables in its generalized form is given below. Here X ! x , (I 1.4f)
x_ n ¼ f n ðx; u and tÞ
Y !
y A, B, C, D and U !
u in capital letters have been i.e. x_1 ¼ a11 x1 þa12 x2 ......a1n xn þ b11 u1 þ b12 u2 ...... þb1r ur
used to mean various matrices detailed discussed later. Two
x_ 2 ¼ a21 x1 þ a22 x2 ...... a2n xn þ b21 u1 þ b22 u2 ......þb2r ur
equations viz. Eqs. I 1.4d and I 1.4e are referred to as state .
equation and output equations (discussed later). x_ n ¼ an1 x1 þ an2 x2 ...... ann xn þ bn1 u1 þ bn2 u2 ...... þbnr ur
X_ ¼ AX þ BU (I 1.4d) (I 1.4g)

FIGURE I 1.4a State space system

explanation.
52 Plant Intelligent Automation and Digital Transformation

2 3 2 3 2 3
2 3 x1 2 3 u1 themselves and the inputs. In order to do the same, it is
x1
_
6 7 a1n 6 7 b1r 6 7 necessary to integrate dx/dt. Number of such integrator is
6 7 6 a11 / 6 7
76 x2 7 6
b11 /
76 7
6 x2
_ 7 6 u2 7 dependent on order number of system. Typical such block
6 7¼6 « 1 « 756 7 þ 6 « 1 « 7
56 7
6 « 7 4 6 « 7 4 6 « 7
4 5 an1 / ann 4 5 bn1 / bnr 4 5 diagram has been shown Fig. I 1.4a2 where, typical linear
xn
_ xn ur first order system has been depicted. In the diagram number
(I 1.4h) of integrators are shown in dotted to indicate that when the
h i order of the system changes there will be changes in the
Or X_ ¼ ½A ½X þ ½B ½U (I 1.4i) number of integrators in the system. For further discussions
refer I 13.3.1 also.
So, Eq. I 1.4i, is the generalized form of Eq. I 1.4d.
1.5 Optimization of controls and
1.4.4 Output equation automations
Outputs stand to represent any system physical variable of
Now let us look into system optimization controls in brief.
interest generated (on account of interaction of the system
through state variables, with inputs). Here state space
(equation) description is that all system variables may be 1.5.0 Basics of process control and process
represented by a linear combination of the state variables xi automation
and the system inputs ui as indicated above. Applying the Let the discussions start with recapitulating discussions on
same principles, one gets Model predictive control (MPC) as depicted in Fig. I 1.3d.
y1 ¼ f 1 ðx; u; tÞ MPC can be extended for process control optimization
(I 1.4j) using Advanced process control (APC), which can be an
yn ¼ f n ðx; u; tÞ umbrella control system to cover a broad range of tech-
niques e.g., fuzzy control/statistical control. When MPC are
i.e. y1 ¼ c11 x1 þc12 x2 ...... c1n xn þd11 u1 þd12 u2 ...... þd1r ur
used with multivariable controller and integrated with an
y2 ¼ c21 x1 þc22 x2 ...... c2n xn þd21 u1 þd22 u2 ...... þd2r ur optimizer, these can bring about lot of benefits for process
Ym ¼ cm1 x1 þcm2 x2 ...... cmn xn þdm1 u1 þdm2 u2 ......þdmr ur control and especially in the applications with dead time
(I 1.4k) dominance, interactions and constraints. As discussed in I
2 3 2 3 1.3.3, in MPC, the effect of past actions of manipulated and
y1 2 3 x1 disturbance variables are taken into account for computing
6 7 c11 / c1n 6 7 the future values of controlled and constraint variables so as
6 y2 7 6 76 7
6 7 6 6 x2 7 to get the desired future profile of controlled variables. So,
6 7¼4 « 1 « 756 « 7
6
6 « 7 7 it is very useful. Therefore, process controls utilizing MPC
4 5 cm1 / cmn 4 5
ym xn with multivariable APCs find their application in Re-
fineries, Petrochemicals, Paper and other Chemical/Power
2 3 plants for optimization. Here, MPC acts as multivariable
2 3 u1
d11 / d1r 6 7 predictive control in APC (ref: I 1.5.2).
6 76
6 u2 7
7
þ6
4 « 1 « 756 7 (I 1.4l)
6 « 7 1.5.1 Structural approach for integrated
dm1 / dmr 4 5
ur optimization and control
Basic purpose of process control is to maintain control
Or; ½Y ¼ ½C ½X þ ½D ½U (I 1.4m) variable and process at desired operating point taking
So, Eq. I 1.4m, is the generalized form of Eq. I 1.4e. into account various process constraints as well as
Here matrix [D] with constant coefficient di,j indicates the ensuring plant reliability and safety for human, process/
weights of input and in many cases [D] is a null matrix. equipment and environment. Whereas optimal controls
help operations to produce quality products at higher
Then; output½Y ¼ ½C ½X i.e.Y ¼ CX (I 1.4n) production rate with minimized costs, and respond to
changing business conditions [20]. So, APC with
multivariable controllers can be used to extend the fa-
1.4.5 Block diagram representation of state cility for Real time optimization (RTO) to improve
profitability and efficiency of the plant by maintaining a
space
process at desired operating point under various process
The matrix-based state equations express the derivatives of limits and constraints. Following are the major steps for
the state-variables explicitly in terms of the states process control with optimization method:
 General discussions on control systems Chapter | 1 53

1. Identification: Identification of the process variable 1.5.2 Discussions on integrated structure

(s) in the question.
Let the discussions start with understanding of nature of
2. Selection: Selection of objective (s) of the functions for
Optimizer, MPC. Subsequently the integrated operation
these.
will be understood.
3. Constraints: Listing out the process constraints.
4. Process Model: Refer I 1.3.0 for process model. 1. Optimizer: Basic optimizers are devised mainly look-
5. Computation of optimization: Optimization involves ing at steady state conditions of the plant, without hav-
realization of the highest possible responsiveness of ing much information on plant dynamics. On the other
the production to market needs, requiring integration hand, because of disturbances a plant can hardly in
across the control hierarchy layers (involving various steady state condition. So, dynamic optimization is a
tasks), allowing both high performance quality control complex process and in integrated system such complex
and flexible transient operation of a process on all optimizer is necessary to handle limits and constraints.
time scales [21]. 2. MPC: Working of MPC as seen in I 1.3.0, is basically
6. Application and Integration: Application of process meant for a particular operating point assuming piece-
constraints developed model and modification (if any). wise linear model across a specific operating point.
7. Simplification: Simplification of models and objectives 3. Integration: As stated above optimization is based on
are necessary to make it workable. steady state model (may be nonlinear) that is assumed
to reflect dominant process mechanisms over the com-
Therefore, it is clear that complete system involves a
plete operating envelope [21]. Therefore, there would
time-scale range from seconds to days at different layers.
be inconsistency between the models applied for opti-
Main objectives of RTO usually include maximize a
mization and MPC. After determination of optimum
product value, production rate and efficiency with focus to
operating conditions by the optimizer, these conditions
minimize a raw material cost downtime etc. Naturally this
are to be realized by MPC. So, MPC is supervised by
requires more complex control strategy involving several
the optimizer as shown. Naturally it is necessary to
manipulated or controlled variables, when compared with
have complex dynamic optimization technique to drive
typical process control. Such integration involves a wide
the plant toward optimum operating conditions
range of process scale at different layers. Also, integration
involving different time scale and process ranges.
must include suitable ways and means for continuous
MPCs to be developed in such a way that, it is consis-
performance diagnosis with sole aim to improve the per-
tent with the models applied by the complex dynamic
formance. Thus, on account of different complexities
plant optimizer and can handle multi-variables. Lying
involved, there exists different strategies at different levels
at higher level, it is the duty of optimizer to set and up-
as shown in self-explanatory Fig. I 1.5a [20]. In terms of
date the steady state targets for MPC for it to execute in
complexity, time of response of RTO lies at higher level
each scan cycle. Optimizer gets limits and constraints
(level 4). It is worth noting that level 3A is a normal
from lower levels to set unconstraint target as long as
regulating loop, whereas level 3B is multivariable (and
there exist a solution. Therefore, MPC working on un-
constraint control) APC needed to shift the operating target
constraint algorithm actually works within the
toward the available limit based on optimum solution
constraint limits [4]. Such realizations are possible in in-
computed at level 4. Level 5 is concerned with planning
tegrated optimization method (ref: Fig. I 1.5.0-a and b).
and scheduling operations for the entire plant [20].
Now let us focus on APC for optimization.
Various constraints and limits are set from lower level
to higher level while based on these set targets emerges
from higher level toward lower level to get the optimized Steady state optimizer
conditions. These have been depicted in Fig. I 1.5b.

MPC
Level 5 Planning and Schedule (days–months)
Level 4 Real time Optimization ( Hrs–days)
DCS
Level 3B Multivariable control & constraints (Min-hrs)
Level 3A Regulatory control (Secs and minutes)
Level 2 Safety: Human, Property & Environment (< 1 sec) Plant Process
Level 1 Field Sensor and Actuator (< 1 sec)
DOWN ARROW UP ARROW
Developed based on [20] TARGET LIMIT
FIGURE I 1.5a Hierarchical structure of control strategy. FIGURE I 1.5b Real time optimizer and control loop.
54 Plant Intelligent Automation and Digital Transformation

1.5.3 Advanced process controller (APC) and l Understanding sustainability methods [23].
optimization l Easy to use model with tuning with minimum
maintenance
Basic control solutions may not be enough for the processes 2. On line optimization: For online optimization continu-
involving complex energy and resource-intensive pro- ously monitor the process to know dynamics of the pro-
cesses. These processes require control systems that can cess and tries to find predicted an optimum operation
maximize efficiency and optimize consumption of energy path through modeling discussed above.
and materials even at the worst condition. APC and opti- l Optimization software packages: They mainly cover:
mizers are two important components here. l Model building application from first-principles

1. Advanced Process Control (APC): APC is a software with solution to various equations with multiple
tool meant to collect process information and apply the variables including finding solutions for nonlinear
right control strategy according to process requirements process.
to determine optimized set points for the most efficient l Carries out “what if” and other analysis methods

performance of the system. APC has the capability to to arrive at optimum points
work in multiple variable environments with constraints l Online reconciliation of data related to all processes

and resolve conflicting process objectives [22]. It can from raw material stage to finished product range.
quickly analyze and adapt to compensate for process With this, discussions on control system are concluded
changes through complicated calculations and adjust- to touch upon basic networking.
ments. So, MPC must create the process dynamics
accurately. APCs are developed with focus on improved
economics of the operation or production. Thus, APC List of abbreviations used
and RTO can improve profitability and efficiency of A/M Auto manual
the plant by accurately operating the plant at desired AI/A/O Analog input/Analog Output
operating conditions under process constraints. ALU Arithmetic logic unit
l APC software: This may include: APC Advanced process control
l Model building (ON/OFF line) verification and API Application programming interface
analysis tool with multiple variables [23]. ARW Antireset windup
l Process historian system to predict future MVs ASIC Application-specific integrated circuit (microchip)
and CVs ATRS Automatic turbine run up system
BMS Burner management system
l Operator guidance for plant operation
BPCS Basic plant control system
l Auto tuning to adapt changes in process dynamics
BTG Boiler turbine and generator
l MPC based APC advantages and disadvantages:
C&I Control and Instrumentation
l Easy to use with process diagnostic features and
CLCS Closed loop control system
easy to maintain CMOS Complementary metal eoxide semiconductor
l Capable of handling graphical interface and CPU Central processing unit
generic equations CV Control valve/Control variable
l Adaptable to diverse control strategy DCS Digital control system/Distributed control system
l Live trending for performance versus constraints DI/DO Digital input/Digital output
[22]. DP (T) Differential pressure (Transmitter)
l APC implementation is costly and time DSP Digital signal processing
ECL Emitter coupled logic
consuming
FCE Final Control element
l Long steps to follow for preconditioning, model
FPGA Field programmable gate array
building, verification, integration and HMI Human machine interface
commissioning HPU Hydraulic power unit
l APC selection: A few selection criteria are: I&C Instrumentation and control
l Application, scalability for multi-process applica- I/O Input/Output
tion with industry focus I/P Input
l New technology with engineering support and ICMP Internet control message protocol
training requirements IIoT Industrial Internet of thing
l Capability for process integration and process IMC Internal model control
identification tool IoT Internet of thing
 General discussions on control systems Chapter | 1 55

IP Internet protocol/Performance index (adaptive control) blogs/controltalkblog/how-do-you-effectively-use-anti-reset-windup-

IR Instruction register options-tips/.
IS Information system or Intrinsic safety [9] L. Rundqwist, Anti-reset wind up for PID controllers, in: IFAC Pro-
IT Information technology ceeding Volumes, vol. 3; Issue 8, Science Direct–Elsevier, 1990 part 4.
MBC/MBPC Model based control/Model based predictive controller [10] The Six: Fundamental Control strategies every process control
MCU Microcontroller (unit) developer must Know; ISA Presentation; https://www.isa.org/
MFR (T) Master fuel relay (trip) WorkArea/DownloadAsset.aspx?id¼153255.
MIS Management information system [11] Override Control; Module 5 Lecture 3; National Programme on
MPC Model predictive control Technology Enhanced (NPTEL); IIT Gawahati & Madras, 2013.
MPU Microprocessor (unit) https://nptel.ac.in/courses/103103037/31.
MRAC Model Reference Adaptive Control [12] T.E. Marlin, Process Control: Designing Processes and Control
MV Measured/Manipulated variable or Medium voltage Systems for Dynamic Performance, second ed., McMaster Univer-
NPSH Net positive Suction Head sity; McGraw-Hill, 2014. http://pc-textbook.mcmaster.ca/Marlin-
O/P Output EntireBook.pdf.
OLCS Open loop control system [13] L. Rogelio, Adaptive Control; Chapter 1 Introduction, Springer,
OS Operating system 1998. https://www.springer.com/cda/content/document/cda./97808
OT Operational technology/Once through (boiler) 57296634-c1.pdf?.
PB Proportional band [14] Concepts, Methods and Techniques in Adaptive Control; G. A.
PC Program counter/Personal computer Dumont and M. Huzmezan; University of British Columbia; https://
PLC Programmable logic controller www.uiam.sk/wslapo/junkyard/acc2002paper.pdf.
PMBC Process model base control [15] P. Jain, D.M.J. Nigam, I.I.T. Roorkee, Design of a model reference
RTO Real time optimization/optimizer adaptive controller using modified MIT rule for a second order
SIL Safety integrity level system, India; Advance in Electronic and Electric Engineering 3 (4)
SIS Safety instrumentation system (2013). https://www.ripublication.com/aeee/62_pp%20%20%20477-
SNMP Simple network management protocol 484.pdf.
SoC System on chip [16] Simple, model-based process control; R. R. Rhinehart; ; Process
SP Stack pointer/Setpoint Automation March/April 2013; (tutorial session at the ISA 2012
TTL Transistor-Transistor logic Automation Week conference);.
WRT With respect to [17] P.E. Orukpe, Basics of model predictive control, in: Supervisor: Dr
I.Jaimuokha, Imperial College, London, 2005. http://citeseerx.ist.psu.
edu/viewdoc/download?doi¼10.1.1.177.4756&rep¼rep1&type¼pdf.
References [18] D.E. Seborg, T.F. Edgar, D.A. Millichamp, Process Dynamics and
[1] S. Bhanot, Process ControlePrinciples and Applications, Oxford Control, second ed., John Wiley and sons Inc, 2004.
University Press, 2010. [19] Internal Model Control (IMC); J. Swevers; Internet Document, July
[2] J.L. Hilburn, P.M. Julich, Microcomputers/Microprocessors: Hard- 2006. http://een.iust.ac.ir/profs/Shamaghdari/MPC/Lectures/lecture_
ware, Software and Applications, Prentice- Hall of India Private IMC.pdf.
limited, New Delhi, 1981. [20] A.A. AMIN, Advanced Process Control and Real-Time Optimiza-
[3] S. Bennett, History of automatic control to 1960: an overview. IFAC tion, Control Engineering, 2016, in: https://www.controleng.com/
Proceedings; Science Direct, 1996. https://www.sciencedirect.com/ articles/advanced-process-control-and-real-time-optimization/.
science/article/pii/S1474667017581355. [21] T. Backx, O. Bosgra, W. Marquardt, Integration of model predictive
[4] B.G. Liptak (Ed.), Instrument Engineers’ Handbook, Volume Two: control and optimization of processes, in: Technical Report
Process Control and Optimization, CRC Press, 2005. LPTe2000e24; IFAC Symposium “Advanced Control of Chemical
[5] Controllers and Controlled SystemsdPart 1; Technical Information: Processes“, ADCHEM 2000, Pisa, Italy, 14e16.6.2000, June 2000.
Samson AG; https://www.samson.de/document/l102en.pdf. https://pdfs.semanticscholar.org/9fd7/9d2d8e7cb32058c968b43c111
[6] G. McMillan, When and How to Use Pulse Width Modulation of defe263887c.pdf.
Controller Outputs, 2017. https://www.controlglobal.com/blogs/ [22] Advanced process control & optimization; Schneider Electric;
controltalkblog/when-and-how-to-use-pulse-width-modulation-of- Internet Document; https://www.schneider-electric.com.my/en/work/
controller-outputs/. solutions/for-business/s4/cement-advanced-process-control-and-opti
[7] Use of anti-reset windup in integral control based learning and re- mization/.
petitive control; Y.S. Ryu and R.W. Longman; IEEE Xplore Digital [23] P. Reynolds, What Is Advanced Process Control (APC) & Online
Library; https://ieeexplore.ieee.org/document/400266. Optimization? ARC Advisory Group, December 26, 2017, in: https://
[8] G. McMillan, How Do You Effectively Use Anti-reset Windup www.arcweb.com/blog/what-advanced-process-control-apc-online-
Options Tips? Control global, 2012. https://www.controlglobal.com/ optimization.
56 Plant Intelligent Automation and Digital Transformation

[24] A history automatic control; C.C. Bissel; In: Nof, S.Y. ed. Springer Further reading
handbook of automation. Springer handbook series (LXXVI). Hei-
delberg, Germany: Springer Verlag, 26 Power plant instrumentation [1] R. Salliss, Optimize your process with advanced process control.
and control Handbook; rev 2; Swapan Basu & A.K. Debnath; Knowledge: Honeywell; Honeywell Users Group Europe, Middle East
Elsevier B.V.; https://www.elsevier.com/books/power-plant- and Africa, 2014, in: https://www.honeywellprocess.com/library/news-
instrumentation-and-control-handbook/basu/978-0-08-102804-9. and-events/presentations/Hon-EMEA14-Advanced-Process-Control.
pdf.